Search

This paper assesses trending AI foundation models, especially emerging computer vision foundation models and their performance in natural landscape feature segmentation. While the term foundation model has quickly garnered interest from the geospatial domain, its definition remains vague. Hence, this paper will first introduce AI foundation models and their defining characteristics. Built upon the tremendous success achieved by Large Language Models (LLMs) as the foundation models for language tasks, this paper discusses the challenges of building foundation models for geospatial artificial intelligence (GeoAI) vision tasks. To evaluate the performance of large AI vision models, especially Meta’s Segment Anything Model (SAM), we implemented different instance segmentation pipelines that minimize the changes to SAM to leverage its power as a foundation model. A series of prompt strategies were developed to test SAM’s performance regarding its theoretical upper bound of predictive accuracy, zero-shot performance, and domain adaptability through fine-tuning. The analysis used two permafrost feature datasets, ice-wedge polygons and retrogressive thaw slumps because (1) these landform features are more challenging to segment than man-made features due to their complicated formation mechanisms, diverse forms, and vague boundaries; (2) their presence and changes are important indicators for Arctic warming and climate change. The results show that although promising, SAM still has room for improvement to support AI-augmented terrain mapping. The spatial and domain generalizability of this finding is further validated using a more general dataset EuroCrops for agricultural field mapping. Finally, we discuss future research directions that strengthen SAM’s applicability in challenging geospatial domains.

The accidental spillage of flammable liquids on in-service ships and offshore installations may lead to pool fires, which are likely to spread over a particularly large area in large compartments under ship motion, resulting in extensive damage. However, the effect of the spreading extent of liquid fuel due to inclined ship motion on pool fire consequences has not been considered in the existing literature. Thus, in this study, fuel discharge experiments were conducted to investigate the spreading behaviour under different substrate inclination angles and discharge rates. The experimental results were analysed to derive closed-form expressions to predict the spreading extent of liquid fuel in large compartments. Additionally, the effects of surface inclination on fire consequences were investigated using the Fire Dynamics Simulator in terms of the heat release rate. The findings can provide guidance for effective fire safety design and establishing a realistic fire modelling methodology for ships and offshore installations.

Phage display is one of the most frequently used platform technologies utilized to screen and select therapeutic antibodies, and has contributed to the development of more than 10 therapeutic antibodies used in the clinic. Despite advantages like efficiency and low cost, it has intrinsic technical limitations, such as the asymmetrical amplification of the library after each round of biopanning, which is regarded as a reason for it yielding a very limited number of antigen binders. In this study, we developed a high-throughput single-clonal screening system comprised of fluorescence immunoassays and a laser-driven clonal DNA retrieval system using microchip technology. Using this system, from a single-chain variable fragment (scFv) library displayed on phages with a complexity of 5.21 × 105 harboring random mutations at five amino acid residues, more than 70,000 clones—corresponding to ~14% of the library complexity—were screened, resulting in 78 antigen-reactive scFv sequences with mutations restricted to the randomized residues. Our results demonstrate that this system can significantly reduce the number of biopanning rounds, or even eliminate the need for this process for libraries with lower complexity, providing an opportunity to obtain more diverse clones from the library.

Comparisons between bin and bulk cloud microphysics schemes are conducted by simulating a heavy precipitation case using a bin microphysics scheme and four double-moment bulk microphysics schemes in the Weather Research and Forecasting (WRF) model. For this, we implemented an updated bin microphysics scheme in the WRF model. All of the microphysics schemes underestimate observed strong precipitation, but the bin microphysics scheme yields the result that is closest to observations. The differences among the schemes are more pronounced in terms of hydrometeor number concentration than in terms of hydrometeor mixing ratio. In this case, the bin scheme exhibits remarkably more latent heat release by deposition and riming than the bulk schemes. This causes stronger updrafts and more upward transport of water vapor, which leads to more deposition, and again, increases the latent heat release. An additional simulation using the bin scheme but excluding the riming of cloud droplets on ice crystals, which is not or poorly treated in the examined bulk schemes, shows that surface precipitation is slightly weakened and moved farther downwind compared to that of the control simulation. This implies that the more appropriate representation of microphysical processes in the bin microphysics scheme contributes to the more accurate prediction of precipitation in this case.

Aerosol-cloud-precipitation interactions in deep convective clouds are investigated through numerical simulations of a heavy precipitation event over South Korea on 15⁻16 July 2017. The Weather Research and Forecasting model with a bin microphysics scheme is used, and various aerosol number concentrations in the range N0 = 50⁻12,800 cm−3 are considered. Precipitation amount changes non-monotonically with increasing aerosol loading, with a maximum near a moderate aerosol loading (N0 = 800 cm−3). Up to this optimal value, an increase in aerosol number concentration results in a greater quantity of small droplets formed by nucleation, increasing the number of ice crystals. Ice crystals grow into snow particles through deposition and riming, leading to enhanced melting and precipitation. Beyond the optimal value, a greater aerosol loading enhances generation of ice crystals while the overall growth of ice hydrometeors through deposition stagnates. Subsequently, the riming rate decreases because of the smaller size of snow particles and supercooled drops, leading to a decrease in ice melting and a slight suppression of precipitation. As aerosol loading increases, cold pool and low-level convergence strengthen monotonically, but cloud development is more strongly affected by latent heating and convection within the system that is non-monotonically reinforced.